Experiment Analysis: Controller parameter stability search

Perform shocks of ETH price to test controller parameter stability, without stochastic processes.

Setup and Dependencies

Load Results

Using the experiment logs, select the experiment of interest from the specific HDF5 store file (these datasets are very large, and won't be committed to repo):

Get experiment exceptions, tracebacks, and simulation metadata for further analysis:

Post Process Results

Remove substeps, add set_params to dataframe, and add post-processing columns:

Control Parameters

Simulation Analysis

Get the initial target price to test stability conditions:

Find all controller constant subsets where the price goes to zero:

Find all controller constant subsets where the price goes to infinity:

Create dataframe of stable simulation scenarios.

Stability is defined as:

  1. The market price and scaled target price remaining within 0.1x and 10x the starting price, for all timesteps